Sequence_overlap not being applied correctly in pose classification #22
Description
in tao_triton/python/utils/pose_cls_dataset_convert.py
step is calculated as shown below.
step = int(sequence_length * sequence_overlap)
In the default settings .yaml file, it is 0.5, which is fine, but for 0.4 or 0.6, the overlap bands are reversed.
For example, if sequence_length is 100 and sequence_overlap is 0.4, step will be 40, the first element of pose_sequences will be 0 to 99 pose sequences, and the second element will be 40 to 139. This means that the actual overlap will be 0.6, not 0.4 (40%).
Is this what is intended?